AI-driven flash shortage forces storage rethinking

AI-driven flash scarcity has emerged as one of the defining infrastructure challenges in the machine learning boom. More than simply waiting until the end of a typical hardware cycle solves the problem, this shortage is forcing a bigger rethink in how companies utilize their existing capacity.

Unlike past shortages, this one is hitting even the biggest buyers. Phil Manes (pictured, right), vice president of go-to-market execution at Vast Data Inc., said current patterns suggest the shortage is unlikely to be resolved within normal product cycle timelines.

“We’re seeing some of the largest customers on the planet trying to grow as usual and at the same time bring all these new applications and AI into their environments and not getting the allocation they need,” Manes said. “This is serious and is expected to be in full swing throughout this year and possibly into the middle of next year.”

Manes and Ace Stryker (left), director of AI and ecosystem marketing for Solidigm, a trademark of SK Hynix NAND Products Solutions Corp., spoke with theCUBE’s Dave Vellante and Rebecca Knight at Vast Forward 2026 during an exclusive broadcast on SiliconANGLE Media’s livestreaming studio, theCUBE. They discussed the growing AI-driven flash shortage and the need for more efficient storage architectures to address it. (*Disclosure below.)

AI-driven flash shortage demands efficiency

AI systems are increasingly used not only by humans but also by agents and automated systems that continuously make requests through application programming interfaces. According to Stryker, the explosion in machine-to-machine traffic is changing the way computing and memory resources are consumed.

“These models have context windows and are increasingly growing, longer loops, more iterations in specific interactions with the model,” he said. “All of this has an incredible impact on storage. And this doesn’t seem to be cyclical and looks like it’s likely to wane soon. That’s the situation we find ourselves in in 2026.”

With flash supplies limited, the focus has shifted to architectural efficiency, specifically how inference workloads consume memory and storage resources. According to Stryker, expanded context windows and more iterative model interactions make it easier to access large amounts of active data, and the main focus these days has been on inference context and key-value caching. This emphasis on efficiency also shapes how Solidigm works with its partners and customers, including Vast, an enterprise data platform provider focused on AI-optimized storage infrastructure.

“[The answer is] “Freeing up space and doing more with less has been the name of the game in AI outcomes for the past few years, with an overall focus on efficiency,” Stryker explained. I can’t find a way out of this. ”

Stay tuned for the full video interview as part of SiliconANGLE and theCUBE’s Vast Forward 2026 coverage.

(*Disclosure: TheCUBE is a media partner of Vast Forward. Sponsors of theCUBE’s coverage, including presentation sponsor Solidigm, have no editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE